Abstract: The key point of realizing self-sensing active magnetic bearing is to extract rotor displacement information from coil current signal accurately. Affected by the variation of switching power amplifier’s pulse width modulation（PWM） duty cycle and rotor position, the coil current has complex and time-varying frequency spectrum. Former analysis and compensation methods of the switching harmonics demodulation type self-sensing active magnetic bearings are mainly based on a completely ideal rotor displacement demodulator model, so that it is difficult to precisely analyze the real demodulator’s output property. By utilizing the cosine Fourier series of the absolute value function and Jacobi-Anger identity, we established the frequency domain analytical model of each part of the self-sensing demodulator under static and dynamic coil current. Then the model is validated by experiment on a 4-DOF radial active magnetic bearing rigid rotor system platform. It is shown that the output of the demodulator under static coil current is the constant value which is proportional to air gap of active magnetic bearing. While the demodulator’s output under dynamic coil current includes three components, whose amplitude and distribution in frequency domain depend on the rotor position and the PWM duty cycle’s change of power amplifier.